ollama-agent-queue

Serialize local Ollama model calls using a queue manager to ensure only one invocation runs at a time, preventing VRAM saturation and stalls across multiple skills.

Serialize concurrent Ollama agent invocations by enqueuing requests into a queue.json file and using a queue.lock to enforce one-at-a-time execution across skills.

Retrieve queued local model call results by polling structured callback files written by the queue manager after the serialized execution completes.

Pause and resume a queue of Ollama agent calls using built-in status commands, which also support clearing the queue and handling offline Ollama outages.

If Ollama goes offline while processing the queue, the queue manager handles the outage through timeout and error handling, keeping queued requests intact for later resumption.

Queue serialization works with multiple skills calling the local model simultaneously by coordinating cross-skill requests, ensuring sequential execution via a lock file and callback polling.